That sounds exactly right to me, or at least this is one specific way to implement a solution to the problem. Good question on size. Speculating… I would guess the size of the index relates not to the size of the model, but to the size of the training data. If you built an extremely naive and straightforward uncompressed index like this, you could imagine building the complete list of known tokens, and for each one the complete list of pairs of source index and character offset into each source. If a pair of 64 bit ints was sufficient for this, you’d have 128 bits per token for the index, compared to whatever the average size token is (which I’d guess is something like 2-4 bytes uncompressed). If that napkin math is anything close to reality, then I might expect the index to be ~8x the size of the training data (with error bars of like 2x-4x on both sides). Maybe we just call it an order of magnitude?

I think you'd find that most code matches other existing code.

Some code, like the famous fast inverse square root, is widely shared. Even trivial code is often just copy-pasted from a popular SO answer.

Other code is driven by something like convergent evolution. It ends up similar to other code because of the limits of well known algorithms, language syntax, APIs, boilerplate, common coding styles, etc.

In other words, I doubt most human programmers would pass a "copyright check" against all existing open source code.

If true, how do you think these things should reflect on copyright or AI policies? Does it matter whether “most” human written code meets the same standard that we have for AI, and if so, why?

It’s hard to discuss unverifiable claims of “most” code. A lot of code, maybe most, or maybe not, is proprietary and kept within corporate walls, so we have no idea how much is or is not copied. SO answers are allowed to be copied, by definition. (And at least with SO answers, the code is publicly accessible, and attribution tracking after the fact is closer to possible, right?)

You do have a good point that attribution for some code could be practically impossible to track. This might get a lot worse if we allow AI to remix and republish it, that could even cause feedback loops if we’re not more careful about tracking attributions.

Copyright requires an element of creativity and originality. The easiest code snippets for AI to reproduce are things like language syntax (loops, definitions, etc), implementations of common algorithms, and boilerplate, and an index would probably find a lot of matches for those, but are they original or creative?

Incidentally, answers on SO are licensed cc-by-sa, so it's easy to violate copyright when copying them; but no one seems to care, suggesting that small code snippets aren't an important copyright concern.

So the challenge for an index would be finding a rare case of creative, important code in a sea of trivial matches.